Knowing how to clone someone’s voice with AI has never been more accessible — but the harder question, the one most tutorials skip, is whether it is legal and whether it is ethical. This post covers both before it covers the how-to, because the legal landscape shifted meaningfully in 2024 and several people have already faced criminal charges or civil lawsuits for voice cloning done wrong.
If you want to clone your own voice, or use voices you have explicit permission to clone, keep reading — the step-by-step is further down. If you are looking to clone someone else’s voice without asking them, the short answer is: do not do it.
TL;DR
- Voice cloning another person without their explicit consent is illegal in most jurisdictions and getting more illegal fast (Tennessee ELVIS Act 2024, EU AI Act, US Senate NO FAKES Act pending)
- The January 2024 Biden robocall incident is the most visible example of what the legal consequences look like
- Cloning your own voice, a consenting voice actor, or a public-domain historical voice is generally lawful — but disclosure is increasingly required
- The technical how-to is straightforward once consent is sorted: 3–5 minutes of audio, local training, real-time output in under 20 minutes
- VoxBooster runs all training and inference locally — your audio never leaves your PC
- Disclosure, watermarking, and consent documentation are the table-stakes of responsible voice cloning in 2026
Is it legal to clone someone’s voice? The short answer
Voice cloning AI creates a synthetic replica of a person’s vocal identity — their timbre, resonance, accent, and prosodic patterns — using a neural model trained on real recordings. Once trained, that model can say anything in the target person’s voice. That capability, applied without consent, is what regulators worldwide have been scrambling to address since 2022.
The short legal answer: in most jurisdictions, cloning someone’s voice without their permission is already illegal, or actionable enough that you do not want to test it. The long answer involves several overlapping legal frameworks, which vary by country and US state.
The legal landscape in 2026
United States: right of publicity + new AI-specific laws
The US has no single federal voice-cloning law — yet. But protection comes from three directions:
Right of publicity. At least 35 US states have right-of-publicity statutes that protect a person’s name, likeness, and voice from unauthorized commercial exploitation. California’s statute (Civil Code § 3344) and New York’s law are the most litigated. These predate AI, but courts have applied them to voice cloning cases.
FTC rules on impersonation. The Federal Trade Commission’s impersonation rules prohibit using AI-generated voices to impersonate government officials or businesses. In 2024, the FTC expanded its enforcement focus to AI-generated impersonation specifically.
The Tennessee ELVIS Act (2024). The Ensuring Likeness Voice and Image Security Act, signed into law in March 2024, is the first US legislation targeting AI voice cloning directly. It makes it a civil and criminal offense to use AI to reproduce a person’s voice without consent for commercial purposes. The name honors Elvis Presley, but it protects everyone — not just celebrities. Several other states have introduced similar bills.
The NO FAKES Act. A bipartisan US Senate bill (introduced in 2023, reintroduced in 2024) would create a federal right to control AI-generated replicas of a person’s voice, image, or likeness. It has not yet passed as of writing, but its trajectory indicates where federal law is heading.
Political deepfake laws. At least 20 states have laws specifically targeting AI-generated deepfake content in political contexts. Deepfake voice consent is a hard legal requirement for any political content in those states — not just a best practice. The January 2024 Biden New Hampshire robocall — in which a cloned version of President Biden’s voice told Democratic voters not to vote in the primary — led to FCC fines and criminal referrals. That incident is the clearest recent example of what misused clone someone voice AI looks like in practice and what the legal response looks like.
European Union: AI Act + GDPR
The EU AI Act, which entered its phased application in 2024–2025, classifies AI systems used to generate or manipulate synthetic media of real people — including voice deepfakes — as systems requiring disclosure. Content generated by AI that could deceive the public must be labeled. Platforms that distribute AI-generated voice content without labeling face significant fines.
GDPR is separately relevant: a person’s voice is biometric data under Article 9 when processed for identification. Cloning a voice involves processing that data. Without a lawful basis (which, absent consent, is hard to establish), GDPR violations are possible even before any content is published.
International variation
The UK, Canada, Australia, Japan, and South Korea all have either pending or enacted legislation on AI-generated likeness. The direction is consistent: consent and disclosure are becoming legal requirements globally, not just ethical best practices.
Voice cloning legality: what is clearly OK
Before covering the how-to, it helps to be explicit about which use cases are unambiguously fine:
Your own voice. Cloning your own voice — for content creation, accessibility, dubbing, or any other purpose — is fully legal. You hold the rights to your own voice. This is the most common use case for tools like VoxBooster.
A consenting voice actor under contract. Commercial voice cloning with a signed agreement — specifying what the clone will be used for, for how long, and what compensation the voice actor receives — is legal and already standard practice in audiobook production, game development, and animation. SAG-AFTRA has published guidelines for this.
Historical/public-domain voices. Voices of people who died before modern recording — historical figures from the 18th and 19th centuries — do not have right-of-publicity protections in most jurisdictions. Reconstructions of Abraham Lincoln’s or Nikola Tesla’s voice from written records are legally distinct from cloning a living person’s voice.
Licensed pre-built voices. Platforms like ElevenLabs, Resemble.ai, Murf, and VoxBooster’s own library include voices licensed from voice actors who consented and were paid. Using those voices within the platform’s terms of service is lawful.
Fictional characters you created. If you invented a character and recorded a voice for them, cloning that character voice is cloning your own intellectual property.
Voice cloning legality: what is clearly NOT OK
Impersonating a living person to deceive others. Using a cloned voice to make someone believe they are hearing the real person — in a phone call, a voice message, a podcast, a video — without disclosure is the core harm these laws target. This applies whether the target is famous or not.
Fraud and scams. Voice cloning for financial fraud (“grandparent scams,” CEO fraud, wire transfer authorization) is a federal crime under existing wire fraud statutes, independent of any AI-specific law.
Non-consensual likeness in sexual content. Multiple states specifically prohibit AI-generated sexual content using a real person’s voice or likeness without consent. This is a separate criminal exposure beyond right-of-publicity.
Political advertising without disclosure. Using a cloned voice in political advertising without clear disclosure violates laws in at least 20 US states and EU regulations.
Cloning a person’s voice to damage their reputation. Even if the content does not involve fraud or sexual context, using a cloned voice to make a real person appear to say defamatory things is actionable under defamation law independent of AI law.
How to get consent properly
If you want to clone a voice actor’s or collaborator’s voice, a verbal “sure, go ahead” is not sufficient. The consent should be:
- Written and signed. A document (even a short email confirmation with explicit language) stating that the person consents to having their voice cloned for AI synthesis purposes.
- Use-specific. The consent should specify what the clone will be used for, on which platforms, and whether commercial use is included.
- Revocable with a process. The person should know they can withdraw consent and what happens to the model if they do.
- Compensated if commercial. If you profit from the content produced with the clone, the voice actor should be compensated — that’s the direction SAG-AFTRA guidelines and emerging state laws are pushing.
Getting AI voice clone consent right matters both legally and practically. Tools like ElevenLabs have built a structured consent flow into their Voice Capture feature — you upload a consent recording where the person verbally confirms they are consenting to cloning. That is a reasonable template regardless of which tool you use.
Voice clone ethics beyond legality
Law lags behind technology. Something can be legal and still cause harm. Voice clone ethics is a distinct conversation from voice cloning legality — and in fast-moving areas of AI, it is often the more useful one. The ethical considerations worth thinking through:
The listener’s right to know. When you publish content using a cloned voice, the listener generally cannot tell without disclosure. That information asymmetry matters. The practice of disclosing AI-generated voices — in credits, in descriptions, in on-screen labels — is emerging as a baseline norm, and the EU AI Act is beginning to codify it.
Consent is ongoing. A voice actor might consent to one project. Repurposing the model for new content without asking again is an ethical problem even if the original consent was documented.
Power asymmetry. It is much easier to clone someone’s voice without their knowledge than it is for that person to detect and stop you. Recognizing that asymmetry — and choosing not to exploit it — is the ethical choice.
Synthetic media transparency. Organizations like the Partnership on AI and initiatives like C2PA (Content Credentials) are building technical standards for labeling AI-generated audio. Embedding those credentials in content you produce is fast becoming standard practice.
What voice cloning actually is (technically)
Understanding the technology helps clarify the risks. There are two main approaches:
RVC (Retrieval-based Voice Conversion). The dominant method for real-time use. RVC trains a model on your target voice samples, then at inference time converts your incoming speech — phoneme by phoneme — into the target voice’s timbre. The model does not generate speech from scratch; it re-synthesizes your speech in the target’s voice. This is what VoxBooster and most real-time tools use.
Neural TTS (text-to-speech). A separate text string is passed to a model that generates speech in the target voice. ElevenLabs, Murf, and PlayHT primarily work this way. The output can be high quality, but it requires typing input rather than speaking naturally. Not suitable for real-time conversation.
Both methods require training data — recordings of the target voice. RVC-based tools (the real-time voice clone vs voice effects comparison covers why RVC dominates for live use) can produce acceptable results from 30 seconds of audio. Neural TTS typically requires more data for good results. Sample size required: 30 seconds (functional) to 5 minutes (good quality) for RVC; 15–30 minutes for high-quality neural TTS clones.
Step-by-step: how to clone someone’s voice with AI legally
This section assumes you are cloning either (a) your own voice, or (b) a voice for which you have written consent. Do not follow these steps for anyone else’s voice.
Option A: Clone your own voice with VoxBooster
VoxBooster runs all training and processing locally on your Windows PC. Your audio never leaves your machine — an important consideration if privacy matters to you.
What you need:
- Windows 10 or 11, 64-bit
- A decent microphone (dynamic or condenser)
- A quiet room for recording
- VoxBooster installed (3-day trial, no card required)
Step 1: Record your reference audio.
Open VoxBooster, go to Voice Clone → My Voice → Create new model. The recording wizard prompts you to speak naturally for 3–5 minutes. Read an article or describe something in your own words — you want natural intonation variation, not a monotone recitation. AC off, windows closed, mic about 5 inches from your face.
Step 2: Review the cleaned audio.
VoxBooster runs automatic noise reduction on the recording before training. Listen to the preview. If there are artifacts or heavy background noise, re-record; it takes five minutes and the model quality difference is significant.
Step 3: Train the model.
Click Train. On an NVIDIA RTX 3060 or better, training 5 minutes of audio takes 10–15 minutes. On an older GPU or CPU-only system, 20–40 minutes. You can leave it running in the background.
Step 4: Test and use.
When training finishes, select your custom model from the list, enable Real-time, and speak into your microphone. Your cloned voice outputs live — usable in Discord, streaming, calls, or any app that reads a microphone input. Read the VoxBooster voice clone tutorial for full detail on each step.
Option B: Clone a consenting voice actor’s voice
The technical process is identical to Option A. The difference is that you do the reference recording session with the voice actor, using their voice, and you have signed consent documentation in hand before you start.
Practical notes:
- Record in a treated space (not a bathroom, not an open office)
- Use the highest quality microphone available — the model’s ceiling is set by the input quality
- Cover a range of speech: questions, statements, fast speech, slow speech, emotional range
- Keep the raw recording files archived alongside the consent documentation
Option C: Use a pre-licensed voice from a library
Both VoxBooster and tools like Murf, Voice.ai, and Resemble.ai include pre-built voices licensed from consenting voice actors. Using those within the platform’s terms of service is the simplest legal path if you need a non-self voice for content.
VoxBooster’s library is accessible from the Voice Clone tab — select a voice, enable Real-time, done. No training required, no recording needed, full licensing already handled. See pricing for what is included on each plan.
Detection and disclosure best practices
Responsible voice cloning in 2026 includes being transparent about what you produced.
Disclose in credits and descriptions. If a video, podcast, or audio file contains AI-cloned voice, say so. A single line in the description (“voice generated with AI”) is a reasonable minimum.
Use content credentials (C2PA). The Coalition for Content Provenance and Authenticity (C2PA) has published a standard for embedding metadata in audio files that records how they were created. Support is growing across tools and platforms.
Do not use a clone to impersonate the original person in contexts where deception is possible. The line between “this character’s voice was AI-generated” and “this is a real recording of [person]” is the ethical line.
Label political or public-interest content explicitly. If voice-cloned content touches on politics, public figures, or matters of public concern, disclosure should be prominent — not buried in credits.
Honest limits of current voice cloning AI
Even the best 2026 tools have failure modes worth knowing:
Strong accents bleed through. If your source voice has a thick regional accent and the target voice does not, the clone will carry traces of your source accent. This is not a bug — the model carries your prosody.
Emotional extremes degrade quality. Models trained on conversational speech perform worse when the input voice is screaming or whispering. Stay in the normal conversational range for best results.
Artifacts on non-native phonemes. If the target voice was trained on English and you speak words in another language, phoneme mismatches introduce artifacts.
The “uncanny valley” at close listening. Voice clones pass casual listening well. Forensic analysis — or a listener who knows the person’s voice intimately — will often detect it. This is partly why disclosure remains the right default even when the quality is high.
Conclusion: clone someone’s voice with AI responsibly
The technical barrier to voice cloning has fallen to near zero. The ethical and legal bar has risen steeply in response. The honest framing of “how to clone someone’s voice” in 2026 is: with consent, with disclosure, and with an understanding of the laws in your jurisdiction.
For the use cases that are clearly safe — your own voice, a consenting collaborator, licensed library voices — the process is straightforward and the results are genuinely useful. VoxBooster makes it accessible on Windows without a cloud subscription or complex setup: download the 3-day trial, record 3–5 minutes of audio, and your local model is ready in under 20 minutes. See the full plan comparison if you decide to continue past the trial.
For everything else: get consent in writing, disclose in your content, and check the laws in your state or country before you publish.
Further reading: Voice Clone vs Voice Effects — which one do you actually want? — Best Voice Changer in 2026 — Best Voicemod Alternative in 2026